5 mistakes when using AI-written ad copy.

Writing for the AI, not the customer

Most operators paste their product brief into ChatGPT, read the output once, and publish it. The copy is grammatically clean and completely inert. It was written to satisfy the prompt, not to land in the mind of a 34-year-old gym owner in Indore scrolling Meta at 9pm on a Tuesday after a 14-hour day.

The fix is not a better prompt. The fix is asking yourself, before you prompt anything: what does this specific buyer already believe, what do they distrust, and what would make them stop scrolling? Write that answer down. Then prompt with it. The AI can execute; it cannot supply the customer insight you have not given it.

Ignoring the platform language

AI defaults to a Twitter-thread voice unless you specify otherwise. That voice works on Twitter. It does not work on Google Search, where someone has already typed their intent and needs a direct answer in 30 characters. It does not work on Meta Reels, where the first frame has to stop a thumb in under two seconds. It does not work on LinkedIn, where a D2C brand talking to distributors needs to sound like a business, not a personal-brand content creator.

Tell the AI the placement before anything else. Not the product, not the audience. The placement. “Write a Google Search headline for a protein brand targeting gym owners in Tier 2 cities” produces completely different copy than “write an ad for my protein brand.” Platform context is not a nice-to-have. It is the first sentence of every good ad brief.

Stripping all specificity

AI loves abstractions. Left alone, it will produce copy full of phrases like “premium quality,” “best-in-class results,” and “transform your business.” These phrases have been used on roughly 40 million ads. They do not convert.

Real ads convert on specifics. “₹1,499 protein, 23g per scoop, lab report attached” is a headline. “Premium quality protein at an unbeatable price” is noise. “Your CA files ITR in 3 clicks, not 3 hours” is a headline. “Streamlined accounting for modern businesses” is noise. Feed the AI your actual numbers, your actual claims, your actual proof. If you don’t have specifics to give it, that is a product positioning problem, not a copy problem.

Skipping the negative test

The lazy A/B test is: run AI variant A against AI variant B. Both came from the same model with similar prompts. They will perform similarly. You learn nothing.

The real test is AI copy against your best-performing existing ad. That ad beat everything else you ran. It has earned its baseline. If the AI copy cannot beat it, the AI copy is not ready to run. “Better than another AI draft” is not the bar. “Better than the ad that currently drives your lowest CPL” is the bar. Most AI copy fails this test on the first pass. Run it anyway and use the loss to improve the brief.

Forgetting the Indian context

AI defaults to US tone, US pricing conventions, and US compliance assumptions. That is a problem across every category. A D2C supplement brand running ads during Navratri needs to know that a significant portion of its audience is fasting and the “eat more protein” angle will not land that week. A fintech running acquisition ads needs to know that BSE/SEBI compliance language is not optional boilerplate. A SaaS targeting SMEs needs to show pricing in ₹, not “starting at $X” with a fine-print conversion.

Tell the AI explicitly: INR pricing, Indian festival calendar, regional language considerations if running regional placements, and whatever compliance obligations your category carries. GST on B2B pricing copy matters. “Free trial” framing for fintech products has regulatory edges. None of this is automatic. Every Indian-context requirement has to be prompted in, or the AI will write a perfectly reasonable ad for a Shopify brand in Austin.